697
Crystallization of membrane proteins
Christian Ostermeier* and Hartmut Michelt
Five new membrane protein structures have been
determined since 1995 using X-ray crystallography: bacterial
light-harvesting complex; bacterial and mitochondrial
cytochrome c oxidases; mitochondrial b c 1 complex; and
(x-hemolysin. These successes are partly based on advances
in the crystallization procedures for integral membrane
proteins. Variation of the size of the detergent micelle and/or
increasing the size of the polar surface of the membrane
protein is the most important route to well-ordered membrane
protein crystals. The use of bicontinuous lipidic cubic phases
also appears to be promising.
Addresses
*Department of Molecular Biophysics and Biochemistry, Yale
University, Bass Center 433, Whitney Avenue, New Haven,
CT 06520-8114, USA; e-mail: [email protected]
tMax-Planck-lnstitut for Biophysik, Abteilung for Molekulare
Membranbiologie, Heinrich-Hoffmann-Strasse '7, 60528
Frankfurt/Main, Germany; e-mail: [email protected]
Current Opinion in Structural Biology 199,7, 7:69,7-,701
http://biomednet.com/elecref/O959440XO0,70069,7
alkyl chains in a belt-like manner. The polar headgroups
of the detergents face the aqueous environment.
Any crystallization strategy has to take into account the
amphipathic nature of the surface of membrane proteins.
Essentially, there are two possibilities for arranging
membrane proteins in the form of 3D crystals [1].
First, one can try to form 2D crystals in the plane of the
membrane, and then stack these membranes in an ordered
way. These crystals are called 'type I'.
Second, one can try to crystallize membrane proteins
within their detergent micelle. The crystal lattice will be
established via polar contacts of the polar surface parts
of the membrane protein extending out of the detergent
micelles. In this case, the crystallization procedures are
very similar to those for soluble proteins. For these
'type II' crystals, detergents with relatively small polar
headgtoups should be used in order not to cover too much
of the membrane proteins polar surface.
© Current Biology Ltd ISSN 0959-440X
Introduction
Up until now, the 3D structures of about 8,000 biological
macromolecules, mainly soluble proteins, have been
determined using X-ray crystallography. During the past
few years, the availability of recombinant DNA technology
to produce and to tailor the protein of interest has
substantially contributed to the exponential growth in the
number of new protein structures published each year.
In contrast, the number of known membrane protein
structures is still below 20. This fact is remarkable
considering that close to 40% of the 6,000 gene products
encoded by the genome of baker's yeast are expected to
be integral membrane proteins. These numbers not only
underscore the importance of membrane proteins, but also
emphasize the enormous biochemical and structural work
that remains to be done in the field of membrane proteins.
Membrane proteins are difficult to handle; the difficulties
reside in the amphipathic nature of their surface. They
possess a hydrophobic surface where they are in contact
with the alkyl chains of the lipids, and they possess a
polar surface where they are in contact with the aqueous
phases on both sides of the membrane or with the polar
headgroups of the lipids. In order to solubilize and to
purify membrane proteins one has to add a vast excess
of detergents--amphiphilic molecules that form micelles
above their critical micellar concentration. The detergent
micelles take up the membrane proteins and cover the
hydrophobic surface of the membrane protein with their
Both types of membrane protein crystals are schematically
represented in Figure 1. Mixed types of crystals appear
to be possible; however, the overwhelming majority of
membrane protein crystals belong to type II.
Since 1985, when the structure of the bacterial photosynthetic reaction center was presented as the first membrane
protein, detailed structures from only seven families of
integral membrane proteins have been published (Table 1;
for reviews, see [2]). The past two years, however,
have seen an enormous increase in the number of
newly solved membrane protein structures, including two
bacterial light-harvesting complexes [3,4], bacterial [5] and
mitochondrial [6,7] cytochrome c oxidase, mitochondrial
bcI complex [8], and oc-hemolysin [9]. All integral membrane proteins crystallized so far arc either pigmented or
eubacterial outer membrane proteins. The latter proteins
contain only 13strands in their membrane-spanning section
and are of extraordinary stability:
We discuss the recent advances in membrane protein
crystallization. Clearly, obtaining well-ordered crystals is
the bottle-neck of membrane protein structure determination. The problem of crystallizing membrane proteins
cannot be reduced to the issue of which screening method
or crystallization set up is to be used. Rather, thorough
biochemical work and intensive protein characterization,
in combination with comprehensive screening for the most
suited detergent, may be the most efficient strategy to
cope with the difficulties of membrane protein crystallization.
698
Biophysical methods
crystals obtained so far have been obtained under quite
standard crystallization conditions. If the biochemist and
the crystal-grower has done his/her job well, data collection
and structure determination of a membrane protein are as
straightforward as for nonmembrane proteins. Currently,
flash cooling of membrane protein crystals is often used
to enhance crystal stability in the X-ray beam [10] or
for trapping reaction intermediates; however, establishing
cryoconditions for membrane protein crystals seems to
be much more difficult than for soluble proteins. This
problem may be due to the presence of detergent micellcs
in membrane protein crystals.
Figure 1
...,"-~'
(a)
~ .
r,
i~--~,
,_t,
(b)
J
,
The two basic types of membrane protein crystals. (a) Type I: stacks
of membranes contain 2D crystalline membrane proteins, which are
then ordered in the third dimension. (b) Type I1: a membrane protein
is crystallized with detergents bound to its hydrophobic surface. The
polar surface portion of the membrane protein is indicated by broken
lines; lipids are indicated by spheres with two alkyl chains attached;
detergents are indicated by squares with one alkyl chain attached.
Reproduced with permission from [1].
T h e first step on the way to the structure of a membrane
protein is to obtain a sufficient amount of pure and
homogeneous protein. T h e second step is to find the
one detergent needed to obtain well-ordered crystals for
crystallization. In fact, this is the most critical step for
crystallization; unfortunatel'> it is also the most error-prone
step. Interestingly, finding the optimal crystallization
conditions does not seem to be a bottle-neck. With respect
to precipitating agents and pH, all membrane protein
Most important: the wet-lab biochemistry
For crystallization trials, up to 100mg of pure protein
must be isolated. Soluble proteins can often be obtained
by overexpression of the gene or cDNA, combined with
the use of affinity tags for detection and purification.
Refolding from inclusion bodies sometimes works well.
Engineering membrane proteins for crystallization is
possible in principle [11] but less helpful, as a sufficient
level of overexpression rarely can be achieved. In all
published cases, membrane protein crystals have been
grown from proteins isolated from natural sources. In
nature, mainly photosynthetic membrane proteins and
those from bioenergetics are abundant, which explains
why these membrane proteins are the best characterized
structurally. T h e majority of membrane proteins in the
cell are present at only very low levels. Up until now,
there seems to be no general way to obtain large quantities
of functional membrane proteins using recombinant D N A
techniques [12].
Recently, however, a strategy for the overproduction of
membrane proteins, which are usually lethal to their
host cells, has been published [13"]. This strategy
involves usage of selected Escherichia coil strains and
the bacteriophage T7 R N A polymerase system for the
the overproduction of a number of membrane proteins.
In these strains, membrane proteins are formed in
large amounts as inclusion bodies. Protein yields in the
range of 100rag per liter of bacterial cell culture have
been reported. Unfortunately, the refolding of membrane
proteins from inclusion bodies is mostly an unsolved
problem and is one of the main challenges for the future.
Table 1
Membrane protein familes for which crystal structures exist.
Membrane protein family
Photosynthetic reaction centers
Porins
Light harvesting complexes
Cytochrome c oxidases
co-hemolysin
Cytochrome b c 1 complex
Prostaglandin H 2 synthase
Resolution (A)
Pigmented
~ sheet
2.3
1.8
2.4
2.7
1.9
3.0
3.5
Yes
No
Yes
Yes
No
Yes
Yes
No
Yes
No
No
Yes
No
No
Crystallizationof membraneproteins Ostermeierand Michel
Detergents: expensive soaps
Since the early years of membrane protein crystallization,
choosing the right detergent has been the key to success.
Well-ordered crystals of the photosynthetic reaction center
from the purple bacterium Rhodopseudomonas viridis could
only be grown using N,N-dimethyl dodecylamine-n-oxide
as detergent. Even use of the decyl homolog did not lead
to crystals. Recent experiences confirm this observation.
T h e cytochrome c oxidases provide illustrative examples.
Crystallization attempts with the cytochrome c oxidasc from bovine beef heart mitochondria continued in
Yoshikawa's laboratory for about twenty years, and crystals
have been obtained in a number of different detergents
[14]; however, only the use of n-decyl-13-D-mahoside
(C10-maltoside)--a mild, well-known d e t e r g e n t - - h a s
yielded well-ordered crystals.
Cytochrome c oxidase from the soil bacterium Paracoccus
denitrificans is another typical example. For the purification and crystallization of the four-subunit complex,
only detergents of the mahoside-type can be used. All
other detergents remove subunits III and IV leaving
an active complex consisting of subunits I and II.
Only n-dodecyl-13-D-maltoside (C12-maltoside) leads to
the formation of well-ordered crystals of the four-subunit
oxidase as an complex with an Fv fragment [15].
Recently, the catalytically active two-subunit complex
could be crystallized, again with the help of an antibody
Fv fragment (see also below) in different detergents.
Originally, crystals were grown using the C12-maltoside,
but these diffracted to only about 8/k. Crystals grown
with hexaethylene glycol monododecyl ether (C12E6)
showed the same poor diffraction quality. With the
Cl0-maltoside, no crystals could be obtained at all.
Recently, the Cll-maltoside became also commercially
available. Crystals grown in this detergent diffract to
better than 2.6~ resolution (C Ostermeier, A Harrenga,
U Ermler, H Michel, unpublished data). Similar crystals can be grown with cyclohexyl-hexyl-[3-D-mahoside
(CYMAL-6), but not with cyclohexyl-pentyl-13-D-maltoside
(CYMAL-5). Cyclohexyl-heptyl-13-I)-mahosidc (CYMAL-7)
is not yet commercially available.
These results show that even small chemical differences
in the detergent can cause essential differences in the
crystallization behaviour of these detergent-membrane
protein complexes. T h e conclusion has to be drawn
that more efforts should be put into screening various
detergents for crystallization than into the variation of
other parameters. A major problem may be the high
costs of many detergents. T h e optimal way to cope
with this hindrance is to purify the protein using a
rather inexpensive detergent such as Triton X-100 or
N,N-dimethyldodecylamine-N-oxide and then to exchange the detergent prior to the crystallization attempts.
One should keep in mind that it may be difficult to
control the completeness of the detergent exchange. In
our opinion, the simplest and most efficient method for
699
a complete exchange is ion exchange chromatography,
or another method in which the membrane protein is
bound to column materials and can be washed with a large
amount of buffer containing the new detergent without
being cluted from the column. Gel filtration or exchange
by uhrafiltration is not recommended if one requires a
complete exchange.
It would be helpful if a continuous set of alkyl chain
lengths were commercially available for many detergent
headgroups, for example, the CxEy-detergents are available only with an even number of C atoms. One should
also keep in mind that mixtures of detergents often may
be useful. Finally, a need still exists for new classes of
detergents.
An alternative to the classic detergents may be the
so-called ' a m p h i p o l s ' - - p o l y m e r s that can potentially keep
membrane proteins in aqueous solution [16"]. These
possess a strongly hydrophilic backbone that is decorated
with hydrophobic sidechains, resulting in an amphiphilic
structure. So far, amphipols have not yet been used for
crystallization, but they might be useful in the future.
Crystallization: finding the needle
(detergent?) in the haystack
For the time being, most promise lies with trying to obtain
a type II crystal. This approach has the advantage that the
membrane protein surrounded by its belt-like detergent
micelle can be treated as an ordinary soluble protein, and
standard crystallization procedures can be used (for general
reviews, see [1,17-20]). Most membrane protein crystals
have been obtained using standard precipitants like
polyethyleneglycols or salts (ammonium sulfate, potassium
phosphate). T h e vapour-diffusion method with sitting
drops is most frequently applied to achieve supersaturation
of membrane proteins.
As outlined above, the choice of the detergent is the most
important factor apart from the stability and homogeneity
of the protein. This is understandable because the
detergent micelle has to fit optimally into the crystal lattice
of the protein. Attractive, polar interactions between
neighboring detergent micelles appear to be helpful
and to contribute to the stability of the crystal lattice.
Such contacts cannot occur when the detergents have
a rather short alkyl chain, thus explaining why crystals
are sometimes obtained only with longer homolog of the
same detergent type. That attractive interactions between
detergent micelles play a role is also indicated by the
fact that crystallization often occurs close to conditions
in which where phase separation into a detergent-rich
and a detergent-depleted phase occurs. This phase
separation is caused by attractive interactions between
detergent micelles [21]. In the case of the bacterial
cytochrome c oxidase crystal, formation normally starts at
the physical boundary between the detergent-rich and the
detergent-depleted phase.
700
Biophysical methods
Detergent micelles can be made smaller by adding
small amphiphilic molecules such as heptane-l,2,3-triol
[1,22,23]. This approach has been successful in the
case of bacterial photosynthetic reaction centers and
light-harvesting complexes [4], for which rather harsh
detergents with small polar headgroups can be used. It
is unsuccessful when rather mild detergents, such as the
alkyl mahosides, are required.
The trick with the complex
Instead of trying to get a smaller detergent micelle, one
can try to increase the surface area of the hydrophilic
portion of the membrane protein. Binding a soluble
protein to the membrane protein under investigation
is one possibility for extending the polar regions. This
strategy has been used successfully twice. The four- and
two-subunit bacterial cytochrome c oxidases have been
crystallized as a cocomplex with an Fv fragment of a
monoclonal antibody ([5]; C Ostermeier, A Harrenga,
U Ermler, H Michel, unpublished data). The crystallization conditions, the space group and the crystal packing of
both complexes are completely different. In both crystal
structures, the Fv fragment plays an essential role in
forming the well-ordered crystal lattice. Another advantage
of using engineered Fv fragments for cocrystallization
is the possibility of using an affinity tag engineered
to the antibody fragment for the rapid isolation of the
whole membrane protein-antibody complex [24]. Thus, an
affinity tag for purification of the membrane protein can be
used even if genetic engineering of the membrane protein
itself is not possible. In the case of the two-subunit oxidase, isolation has been simplified by this strategy. Starting
with crude membranes, crystallization trials can be set up
within six hours after starting purification (C Ostermeier,
A Harrenga, U Ermler, H Michel, unpublished data).
Producing the Fv fragments may be a labour-intensive and
often cumbersome procedure; however, for many important membrane proteins well-characterized hybridoma cell
lines are already available.
arrays spontaneously. In particular, this method appears
to be the only chance for membrane proteins that are
unstable in detergent micelles or in the absence of added
lipids.
Conclusions
The picture that emerges is that the membrane proteins
tend to form the crystal lattice; the custal lattice that
forms is strongly influenced by the polar headgroup of
the detergent. Sometimes, for example in the case of the
photosynthetic reaction center from the purple bacterium
R. viridis, the headgroups are involved by forming critical protein/headgroup/protein contacts (CRD Lancaster,
H Michel, unpublished data). Often, the length of the
alkyl chain of the detergent has to be optimized in order to
get a well-ordered crystal. A possible reason for this is that
polar interactions between neighboring detergent micelles
are needed to stabilize the protein crystal lattice.
The recent advances in structural membrane protein
research raise some hope that crystallography of membrane proteins will be no longer a wallflower in the
field of structural biology but will become a powerful
tool for understanding essential functions of membrane
proteins, such as cell-cell communication via hormones or
neurotransmitters, transport across membranes or energy
conversion. The prerequisite for membrane protein cryst a l l o g r a p h y - m e m b r a n e protein crystallization--is still
far away from being straightforward or routine. Two of the
most important problems to be solved in the near future
arc the overproduction of functional membrane proteins
in their native membrane environments, and the refolding
of recombinant membrane proteins from inclusion bodies.
Patience and many long-term grants are necessary before
we can state that membrane protein crystallography is no
longer in its infancy.
Acknowledgements
\\'e are grateful to Bryan .~utton for reading the manuscript.
Use of bicontinuous lipidic cubic phases
When mixed with aqueous solvents, some lipids form a
bicontinuous cubic phase, in which the lipids are arranged
in a curved, continuous 3D bilayer. Landau and Rosenbusch [25°°] have succeeded to incorporate monomeric
bacteriorhodopsin prepared from purple membranes in
such a bilayer, and to use this as a matrix for crvstallization.
The idea is that the protein can diffuse in the bilayer, but
it is also able to form 3D contacts. Landau and Rosenbusch
have been able to demonstrate that bacteriorhodopsin
forms small, but well-ordered 3D crystals. The X-ray data
obtained from the most well-ordered crystal form indicate
that the same 2D crystal lattice is formed that is observed
in the native purple membrane. These membranes appear
to be stacked and well ordered in the third dimension:
therefore, the crystals belong to type I. It is to be hoped
that this method can also be used for membrane proteins
that do not have a strong tendencv to form 2D crystalline
References and recommended reading
Papers of particular interest, published within the annual period of review,
have been highlighted as:
•
•1.
2.
of special interest
of outstanding interest
Michel H: Crystallization of membrane proteins, Trends Biochem
Sci 1983, 8:56-59.
Walker JE, Saraste M (Eds): Membrane proteins. Curr Opin
Struct Biol 1996, 6:457-510.
3.
McDermott G, Prince SM, Freer AA, Hawthornthwaite-Lawless AM,
Papiz MZ, Cogdell RJ, Isaacs NW: Crystal structure of
an integral membrane light-harvesting complex from
photosynthetic bacteria. Nature 1995, 374:517-521.
4.
Koepke J, Hu X, Muenke C, Schulten K, Michel H: The crystal
structure of the light-harvesting complex II (B800-850) from
Rhodospirillum molischianum. Structure 1996, 4:581-597.
5.
Iwata S, Ostermeier C, Ludwig B, Michel H: Structure at
2.8A resolution of cytochrome c oxidase from Paracoccus
denitrificans. Nature 1995, 376:660-669.
Crystallization of membrane proteins Ostermeier and Michel
6.
TsukiharaT, Aoyama H, Yamashita E, Tomizaki T, Yamaguchi H,
Shinzawa-ltoh K, Nakashima R, Yaona R, Yoshikawa S: Structures
of metal sites of oxidized bovine heart cytochrome c oxidase
at 2.8 A. Science 1995, 269:1069-1074.
7.
TsukiharaT, Aoyama H, Yamashita E, Tomizaki T, Yamaguchi H,
Shinzawa-ltoh K, Nakashima R, Yaono R, Yoshikawa S: The whole
structure of the 13-subunit oxidized cytochrome c oxidase at
2.8A. Science 1996, 272:1136-1144.
8.
Yu C-A, Xia J-Z, Kachurin AM, Yu L, Xia D, Kim H, Deisenhofer J:
Crystal structure of beef heart mitochondrial cytochrome-bc 1
complex. Science 1997, 277:60-66.
9.
Song L, Hobaugh MR, Shustak C, Cheley S, Bayley H, Gouaux JE:
Structure of staphylococcal 0~-hemolysin, a heptameric
transmembrane pore. Science 1996, 274:1859-1866.
10.
Lee JW, Chan M, Law TV, Kwon HJ, .lap BK: Preliminary
cryocrystallographic study of the mitochondrial cytochrome bc 1
complex: improved crystallization and flash-cooling of a large
membrane protein. J Mo/Bio/1995, 252:15-19.
701
16.
•
TribeC, Audebert R, Popot J-L: Amphipols: polymers that keep
membrane proteins soluble in aqueous solutions. Proc Nat/
Acad Sci USA 1996, 93:15047-15050.
A new class of surfactants, the amphipols, is developed. This class of surfactants is able to keep membrane proteins soluble in aqueous solutions.
Because of their size heterogeneity, however, they may not be useful for
crystallization. Further developments may improve these amphipols so that
they can be a useful alternative to detergents in the future.
17.
Garavito RM, Markovic-Housley Z, Jenkins JA: The growth and
characterization of membrane protein crystals. J Cryst Growth
1986, 76:701-709.
18.
Kehlbrandt W: Three-dimensional crystallization of membrane
proteins. Quart Rev Biophys 1988, 21:429-477.
19.
Michel H (Ed): Crystallization of Membrane Proteins. Boca Raton,
FL: CRC Press; 1991.
20.
Garavito RM, Picot D, Loll PJ: Strategies for crystallizing
membrane proteins..I Bioenerg Biomembr 1996, 28:13-27.
21.
ZulaufH: Detergent Phenomena in membrane protein
crystallization. In Crystallization of Membrane Proteins. Edited by
Michel H. Boca Raton, FL: CRC Press; 1991:53-72.
11.
Popot J-L, Saraste M: Engineering membrane proteins. Curt
Opin Biotechnol 1995, 6:394-402.
22.
12.
Griesshammer R, Tate CG: Overexpression of integral
membrane proteins for structural studies. Quart Rev Biophys
1995, 26:315-422.
TimminsPA, Hauk J, Wacker T, Welte W: The influence of
heptane-l,l,3-triol on size and shape of LDAO micelles. FEBS
Lett 1991, 280:115-120.
23.
Gast P, Hemelrijk P, Hoff AJ: Determination of the number
of detergent molecules associated with the reaction
center protein isolated from the photosynthetic bacterium
Rhodopseudomonas viridis. FEBS Lett 1994, 337:39-42.
24.
KleymannG, Osterrneier C, Ludwig B, Skerra A, Michel H:
Engineered Fv fragments as a tool for the one-step purification
of integral multisubunit membrane protein complexes.
Biotechnology 1995, 13:155-160.
13.
o.
MirouxB, Walker JE: Over-production of proteins in Escherichia
coil: mutant hosts that allow synthesis of some membrane
proteins and globular proteins at high levels. J Mo/Bio/1996,
260:289-296.
The overproduction of membraneproteinsin an Escherichia co~i, BL21 (DE3),
T7 RNA polymerase expression system is investigated. Mutant host strains
are selected that produce the protein as inclusion bodies at very high levels.
Seven membrane proteins can be expressed in levels up to 1O0 mg per liter
of culture.
14.
Shinzawa-ltoh K, Ueda H, Yoshikawa S, Aoyama H, Yamashita E,
Tsukihara T: Effects of ethyleneglycol chain length of dodecyl
polyethyleneglycol monoether on the crystallization of bovine
cytochrome c oxidase. J Mo/Bio/1995, 246:5?2-575.
15.
Ostermeier C, Iwata S, Ludwig B, Michel H: Fv fragmentmediated crystallization of the membrane protein bacterial
cytochrome c oxidase. Nat Struct Bio/1995, 2:842-846.
25.
•.
LandauEM, Rosenbusch JP: Lipidic cubic phases: a novel
concept for the crystallization of membrane proteins. Proc Nat/
Acad Sci USA 1996, 93:14532-14535.
Bacteriorhodopsin is crystallized using a system consisting of tipids, which
form bicontinuous cubic bilayer phases, salts, water, and protein. Bacteriorhodopsin prepared in octylglucoside is incorporated into the lipids. The
lipidic phase should act as a sink for the detergent molecules so that the
bacteriorhodopsin molecules diffuse in the membrane system of the lipids
without their detergent micelle. The lipid phases are suggested to provide
nucleation sites and support crystal growth by lateral diffusion of the protein
molecules in the bilayer membrane.